FrontierCode
2026-10-27 · General mergeability
2.0
- What it measures
- General mergeability: would the maintainer actually merge this PR? 2.0 is a 132-task suite. Maintainer-crafted tasks, graded by unit tests, rubrics, and verifiers. Runs flagged for unfair internet use are zeroed. On this site it is a general board and appears in the Web frontend, Android, and CLI / DevOps scenarios.
- How
- Cognition's FrontierCode 2.0 (current revision). The public page defaults to the Main set; this site imports every (model × reasoning effort) row from All reasoning levels, untruncated. The # column ranks models by their best Main effort. The complete JSON includes Obsidian 3 (omitted from the public HTML table); this site keeps the complete JSON. Score is the weighted rubric aggregate; Pass rate, Flag rate, Cost / rollout, and Output tokens match the source columns.
- What the score means
- The percent is the mergeability Score (higher is better). Runs flagged for unfair internet use score 0. Composite treats it as a rate board: ratio to the leader.
- Caveats
- Main and Extended are separate task sets; this page keeps the default Main board only. The public HTML table omits Obsidian 3; the complete data.json leads with Obsidian 3 high ~46.8%, then Claude Opus 5.5 max ~45.5%. Rows such as SWE-* / Beam / Obsidian / Inkling / Mistral without a model page keep the source label only. Version 1.1 stays complete as an archived revision.
- Updates
- Revision 2.0 is current (changelog 2026-10-27). This page synced 102 rows (all reasoning levels for 26 models) from https://cognition.com/data/frontier-code-2.0/data.json on 2026-10-28. The older frontiercode-leaderboard/data.json still only has v1/v1_1 and is not treated as current.
Full board
102
| # | Model | Score | Pass rate % | Flag rate | Cost / rollout | Output tokens |
|---|
| 1 | ?Obsidian 3low | 39.2% | 43.1% | 1.5% | $2.20 | — |
| 1 | ?Obsidian 3medium | 39.8% | 43.7% | 1.5% | $2.84 | — |
| 1 | ?Obsidian 3high | 46.8% | 51.4% | 1.7% | $8.62 | — |
| 1 | ?Obsidian 3xhigh | 45.9% | 50.4% | 2.1% | $19.59 | — |
| 1 | ?Obsidian 3max | 45.5% | 49.9% | 4.5% | $47.44 | — |
| 2 | Claude Opus 5.5low | 14.7% | 16.4% | 1.2% | $0.746 | 14.2k |
| 2 | Claude Opus 5.5medium | 30.1% | 33.0% | 1.5% | $1.70 | 30.1k |
| 2 | Claude Opus 5.5high | 34.5% | 37.9% | 1.5% | $2.42 | 41.2k |
| 2 | Claude Opus 5.5xhigh | 43.4% | 47.3% | 2.1% | $5.18 | 88.1k |
| 2 | Claude Opus 5.5max | 45.5% | 49.7% | 2.9% | $12.82 | 201.9k |
| 3 | GPT-6 Astralow | 31.1% | 34.2% | 0.8% | $2.44 | 9.0k |
| 3 | GPT-6 Astramedium | 36.8% | 40.2% | 1.7% | $3.49 | 14.1k |
| 3 | GPT-6 Astrahigh | 38.3% | 41.4% | 2.4% | $4.28 | 18.4k |
| 3 | GPT-6 Astraxhigh | 41.4% | 44.9% | 1.5% | $4.54 | 20.9k |
| 3 | GPT-6 Astramax | 42.8% | 46.2% | 2.3% | $6.68 | 36.4k |
| 4 | Claude Sonnet 5.5low | 9.4% | 10.6% | 1.1% | $0.369 | 11.3k |
| 4 | Claude Sonnet 5.5medium | 14.2% | 15.8% | 1.4% | $0.482 | 14.4k |
| 4 | Claude Sonnet 5.5high | 21.9% | 24.3% | 1.4% | $1.07 | 27.8k |
| 4 | Claude Sonnet 5.5xhigh | 34.8% | 38.0% | 2.3% | $3.31 | 80.4k |
| 4 | Claude Sonnet 5.5max | 41.0% | 45.0% | 3.8% | $19.94 | 514.7k |